Falcon2
About
The Falcon LLM family, developed by the Technology Innovation Institute (TII), includes a range of large language models (LLMs) with varying parameter sizes, offering diverse capabilities in processing and understanding language 139. The Falcon2 series marks a significant leap forward, prioritizing compact, efficient models with improved performance metrics 21112. Its premier model, Falcon2-11B, features 11 billion parameters and utilizes over 5,000 billion tokens of RefinedWeb data, supplemented by specially curated datasets 257. This model outperforms Meta's Llama 3 8B and rivals Google's Gemma 7B in terms of capability 12. Notably, Falcon2 also includes a Falcon2-11B VLM variant, integrating image comprehension abilities as the only AI model with vision-to-language support 2412. Both models are open-source under a permissive license, designed for efficient operation on a single GPU, with future updates planned to incorporate "Mixture of Experts" to enhance their prowess 12.